基于VMD能量权重法与BWO-SVM的铣刀磨损状态监测

Milling cutter wear monitoring based on VMD energy weighting method and BWO-SVM

针对铣刀磨损状态监测中信号噪声大、监测效率低等问题,提出了一种基于能量权重法的变分模态分解(VMD)与黑寡妇(BWO)-支持向量机(SVM)的铣刀磨损状态监测方法。首先,运用VMD将铣削时产生的振动信号分解成若干固有模态函数(IMF)分量,并通过能量加权合成峭度指标自适应提取出了包含磨损状态特征的IMF分量,并进行了信号重构,对重构信号进行了特征提取;然后,利用BWO算法优化SVM的参数,构建了BWO-SVM铣刀磨损状态监测模型;最后,为了验证上述方法的有效性,以某公司真实加工现场的PHM Society 2010铣刀全寿命周期的振动数据进行了实验,并且又通过实际的工程案例对此进行了验证。研究结果表明:通过所提方法自适应提取有效分量并进行信号重构后,降噪效果明显,并通过与遗传算法(GA)和粒子群算法(PSO)优化的SVM相比,经过BWO优化的SVM的训练时间缩短至25.142 s,同时监测精度达到97.246%;采用该方法对铣刀磨损状态进行监测,能够获得更快的识别速度与更高的准确性,提高了铣刀磨损状态监测的效率。

Aiming at the problem of large signal noise and low monitoring efficiency in milling cutter wear state monitoring,a method of milling cutter wear state monitoring based on the energy weighting method of variational modal decomposition(VMD)and black widow optimization(BWO)-support vector machine(SVM)was proposed.Firstly,VMD was used to decompose the vibration signal generated during milling into a number of inherent modal function(IMF),and the IMF components containing wear state features were adapted to extract the signal reconstruction by energy-weighted synthetic cliff metrics,and features were extracted from the reconstructed signal.Then,the parameters of the SVM were optimized using the BWO algorithm to construct a BWO-SVM milling tool wear state monitoring model.Finally,experiments were carried out with the vibration data of the PHM Society 2010 milling cutter throughout its life cycle and verified by engineering cases.The results show that the proposed method is effective in noise reduction after adaptively extracting the effective components for signal reconstruction,and the training time of the optimized SVM by BWO is shortened to 25.142 s compared with the SVM by genetic algorithm(GA)and particle swarm optimization algorithm(PSO),and the monitoring accuracy reaches 97.246%.The wear condition monitoring of milling cutters by this method can obtain faster recognition speed and higher accuracy,and improves the efficiency of milling cutter wear monitoring.